← sources

ggerganov/llama.cpp b9521: b9521

source: github (github:ggerganov/llama.cpp)published 2026-06-05 06:54ingested 2026-06-05 09:36

retained text

<details open>

CUDA: enroll mul_mat_vec_q_moe into pdl (#24087)

* Enroll mul_mat_vec_q_moe into PDL, boosting MTP performance on BW

Data collected on a B4500:

Before
```
(llama.cpp) ➜  llama.cpp git:(master) ✗ python mtp-bench.py
  code_python        pred= 192 draft= 150 acc= 116 rate=0.773 tok/s=202.8
  code_cpp           pred= 192 draft= 147 acc= 117 rate=0.796 tok/s=212.8
  explain_concept    pred= 192 draft= 161 acc= 110 rate=0.683 tok/s=196.4
  summarize          pred= 192 draft= 138 acc= 122 rate=0.884 tok/s=226.6
  qa_factual         pred= 192 draft= 138 acc= 121 rate=0.877 tok/s=225.1
  translation        pred= 192 draft= 158 acc= 112 rate=0.709 tok/s=201.5
  creative_short     pred= 192 draft= 160 acc= 110 rate=0.688 tok/s=197.2
  stepwise_math      pred= 192 draft= 150 acc= 115 rate=0.767 tok/s=209.2
  long_code_review   pred= 192 draft= 148 acc= 116 rate=0.784 tok/s=208.9
```
After
```
(llama.cpp) ➜  llama.cpp git:(master) ✗ python mtp-bench.py
  code_python        pred= 192 draft= 150 acc= 116 rate=0.773 tok/s=211.9
  code_cpp           pred= 192 draft= 147 acc= 117 rate=0.796 tok/s=224.6
  explain_concept    pred= 192 draft= 161 acc= 110 rate=0.683 tok/s=207.8
  summarize          pred= 192 draft= 138 acc= 122 rate=0.884 tok/s=240.2
  qa_factual         pred= 192 draft= 138 acc= 121 rate=0.877 tok/s=238.5
  translation        pred= 192 draft= 158 acc= 112 rate=0.709 tok/s=213.4
  creative_short     pred= 192 draft= 160 acc= 110 rate=0.688 tok/s=208.8
  stepwise_math      pred= 192 draft= 150 acc= 115 rate=0.767 tok/s=221.7
  long_code_review   pred= 192 draft= 148 acc= 116 rate=0.784 tok/s=220.7
```

Server launched with:
```
➜  llama.cpp git:(osimons/enroll_mul_mat_vec_q_moe_into_PDL) ✗ ./build-x64-linux-gcc-reldbg/bin/llama-server \
    -m /mnt/share/gguf/unsloth/Qwen3.6-35B-A3B-MTP-GGUF/Qwen3.6-35B-A3B-UD-Q4_K_M.gguf -dio \
    --spec-type draft-mtp \
    --spec-draft-n-max 2 \
    -ngl all \
    -fa on \
    --host 0.0.0.0 \
    --port 8080 -np 1 --chat-template-kwargs "{\"preserve_thinking\": true}"
```

* LC to overlap with following kernels

</details>

**macOS/iOS:**
- [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-macos-arm64.tar.gz)
- macOS Apple Silicon (arm64, KleidiAI enabled) [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23780)
- [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-macos-x64.tar.gz)
- [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-xcframework.zip)

**Linux:**
- [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-ubuntu-x64.tar.gz)
- [Ubuntu arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-ubuntu-arm64.tar.gz)
- [Ubuntu s390x (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-ubuntu-s390x.tar.gz)
- [Ubuntu x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-ubuntu-vulkan-x64.tar.gz)
- [Ubuntu arm64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-ubuntu-vulkan-arm64.tar.gz)
- [Ubuntu x64 (ROCm 7.2)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-ubuntu-rocm-7.2-x64.tar.gz)
- [Ubuntu x64 (OpenVINO)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-ubuntu-openvino-2026.0-x64.tar.gz)
- Ubuntu x64 (SYCL FP32) [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23705)

**Android:**
- [Android arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-android-arm64.tar.gz)

**Windows:**
- [Windows x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-win-cpu-x64.zip)
- [Windows arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-win-cpu-arm64.zip)
- [Windows x64 (CUDA 12)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-win-cuda-12.4-x64.zip) - [CUDA 12.4 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/b9521/cudart-llama-bin-win-cuda-12.4-x64.zip)
- [Windows x64 (CUDA 13)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-win-cuda-13.3-x64.zip) - [CUDA 13.3 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/b9521/cudart-llama-bin-win-cuda-13.3-x64.zip)
- [Windows x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-win-vulkan-x64.zip)
- Windows x64 (SYCL) [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23705)
- [Windows x64 (HIP)](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-bin-win-hip-radeon-x64.zip)

**openEuler:**
- [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23705)
- openEuler x86 (310p)
- openEuler x86 (910b, ACL Graph)
- openEuler aarch64 (310p)
- openEuler aarch64 (910b, ACL Graph)

**UI:**
- [UI](https://github.com/ggml-org/llama.cpp/releases/download/b9521/llama-b9521-ui.tar.gz)

document fetched 2026-06-05 09:36